propeller:testing for differences in cell type proportions in single cell data

Abstract

Motivation Single cell RNA Sequencing (scRNA-seq) has rapidly gained popularity over the last few years for profiling the transcriptomes of thousands to millions of single cells. This technology is now being used to analyse experiments with complex designs including biological replication. One question that can be asked from single cell experiments, which has been difficult to directly address with bulk RNA-seq data, is whether the cell type proportions are different between two or more experimental conditions. As well as gene expression changes, the relative depletion or enrichment of a particular cell type can be the functional consequence of disease or treatment. However, cell type proportion estimates from scRNA-seq data are variable and statistical methods that can correctly account for different sources of variability are needed to confidently identify statistically significant shifts in cell type composition between experimental conditions. Results We have developed propeller, a robust and flexible method that leverages biological replication to find statistically significant differences in cell type proportions between groups. Using simulated cell type proportions data we show that propeller performs well under a variety of scenarios. We applied propeller to test for significant changes in cell type proportions related to human heart development, ageing and COVID-19 disease severity. Availability The propeller method is publicly available in the open source speckle R package (https://github.com/phipsonlab/speckle). All the analysis code for the paper is available at the associated analysis website: https://phipsonlab.github.io/propeller-paper-analysis/. The speckle package, analysis scripts and datasets have been deposited at https://doi.org/10.5281/zenodo.7009042. Supplementary information Supplementary data are available at Bioinformatics online.

Single cell DNA methylation ageing in mouse blood

“…The cell type proportions reflect the expected distribution of B-, T, and other blood cells (Fig 1B) (Teo et al, 2021). Examining the cell type proportions as a function of age, using propeller (Phipson et al, 2022), did not show any significant ageing effect. Suggestively, we observe an increase in the B-cells proportion with age, agreeing with prior reports (Teo et al, 2021), but this change was not significant in our sample set, probably due to the lower sample and cell numbers.…”

“…To test for effects of age on cell type composition we used propeller (Phipson et al, 2022), implemented in the Specle R package. We transformed the proportions using the logit function, leveraging “getTranssformedProps()” and used the “propeller.annova()” function to test for the effect of age, and correct for multiple testing.…”

Trellis Single-Cell Screening Reveals Stromal Regulation of Patient-Derived Organoid Drug Responses

“…Second, our PDO-CAF PTM screening data contains >2,500 conditions with >25 million cells. Existing state of the art to analyze such large datasets is to compare cluster proportions between single-cell samples [25–27]. Emerging methods can compare distributions using earth mover’s distance (EMD), but only at course granularity [22], or by using graph diffusion which does not account for the hierarchical tree structure of cytometry data [23].…”

Brain region-specific changes in neurons and glia and dysregulation of dopamine signaling inGrin2amutant mice

“…The propeller function from the speckle R v0.0.3 package (Phipson et al, 2022) was used with arcsin normalization to test if there were statistical differences in major cell type and subtype composition across genotypes within each dataset.…”